(A) study on road parameter estimation using color image for lateral control of vehicles

Abstract

For the autonomous navigation of vehicles in highway, the lateral control is an essential element. Computer vision is the most powerful sensor to detect the road lane or boundary. But the heavy computational load makes it difficult to implement its algorithm in a real-time without using expensive additional hardware. In this thesis, we propose a fast lane search algorithm and the simple and reliable curvature calculation. The search algorithm is an improved version of the conventional boundary following heuristic algorithm in that the proposed algorithm need not have previous edge map. Besides, the shadow treatment by hue is the advantage of the color image. The curvature calculation is made directly from the data set. The algorithm is implemented on a personal computer and the curvature is calculated in a frame rate. The validity of the proposed algorithm is supported by a number of simulations of real road images. The curvature calculated by the proposed algorithm is compared with the curvature obtained from the steering angle while human driver drives the vehicle.